JP7687935B2 - Computer system and device condition search support method - Google Patents

Description

The present invention relates to a system for supporting the search for equipment conditions of manufacturing equipment and a method for supporting the search for equipment conditions of manufacturing equipment.

To increase the number of materials and structures that can be handled , while at the same time significantly improving accuracy , speed, availability, and operating costs, modern control devices are equipped with multiple control knobs (control conditions) to enhance control and manufacturing capabilities.

One method for improving the processing accuracy, processing speed, operating rate, or operating costs of manufacturing equipment is to optimize the equipment conditions. In particular, optimizing the equipment conditions is important for improving processing accuracy and achieving the target processed shape.

To maximize the performance of manufacturing equipment, it is necessary to determine many input parameters that correspond to the many control knobs, i.e., equipment conditions. Also, in order to achieve more difficult processing results, a method is used in which multiple manufacturing processes (steps) are set within a single process and the input parameters are changed for each step. As a result, it is extremely difficult to determine the combination of input parameters that will produce the desired processing results, which lengthens the time it takes to develop processing conditions and increases development and operating costs. Furthermore, there is an increasing demand for highly difficult processing, and there is a shortage of engineers with the advanced knowledge and skills to meet this demand.

For these reasons, there is a demand for a system that supports engineers in optimizing the conditions of manufacturing equipment. Furthermore, a simple and convenient system is required so that engineers without specialized knowledge can use the support functions.

Patent Document 1 discloses a method for optimizing the conditions of semiconductor manufacturing equipment. Patent Document 2 discloses a method for reducing the costs that users spend on simulations by publishing a processing result simulator for semiconductor manufacturing on the Internet.

JP 2019-040984 A JP 2003-50871 A

One way to support engineers would be to provide an optimal condition search system that uses a machining result simulator or a machining shape prediction model that uses machine learning and deep learning.

However, many simulation models and learning models that can handle difficult processing require data scientists and engineers with knowledge of scientific and technical calculations and programming, and require scripts and commands to specify the necessary device conditions and run the simulator, making them difficult to execute. In addition, there are many steps from building the dataset to executing the calculations, and it is often necessary to use multiple software programs in the process. In addition, there is a shortage of engineers with knowledge of scientific and technical calculations and programming.

Therefore, it is necessary to provide a simple and convenient system that can be used by general engineers. Specifically, a system is required that allows one-stop processing from data input to consideration of device conditions, and does not require the creation of commands or scripts.

Since the results of calculations for optimal device conditions are highly dependent on the training data set and calculation settings, it is extremely important to make it easy to change these settings and store the results in order to improve user convenience when searching for device conditions.

In particular, when constructing a learning dataset, it is important to select data from all data or acquired data that can maintain or improve the prediction accuracy of the device conditions using a device condition search model. Therefore, it is important that the learning dataset can be constructed easily.

Patent Document 1 discloses a method for optimizing conditions for semiconductor processing equipment using a learning model. As mentioned above, in order to solve the shortage of engineers in optimizing equipment conditions, such a method needs to be made available to general engineers.

Patent Document 2 also discloses a method for publishing a simulator of the processed shape on the Internet and making it available to multiple users. In order to search for optimal equipment conditions, it is necessary for general engineers to be able to not only predict the shape, but also to incorporate actual experimental data into the prediction and improve the prediction model.

The present invention aims to provide a system that assists in the search for equipment conditions for manufacturing equipment, and in particular, assists in the construction of a learning dataset.

A representative example of the invention disclosed in the present application is as follows: A computer system for assisting in searching for equipment conditions for controlling a manufacturing equipment performing a process, comprising at least one computer having an arithmetic unit, a storage device connected to the arithmetic unit, and a connection interface connected to the arithmetic unit, the process including at least one manufacturing process, the computer system presenting a first interface for receiving input of learning data consisting of a combination of values of input parameters which are the equipment conditions and values of output parameters which are the results of the process, storing the learning data input via the first interface in the storage device, analyzing a structure of the manufacturing process of the process in reference learning data and a structure of the manufacturing process of the process in other learning data, performing a classification process for identifying the learning data which can be aggregated into the same data set as the reference learning data based on the results of the analysis, and aggregating the learning data based on the results of the classification process. a data set is generated by performing a learning process of a model that predicts a value of the output parameter from a value of the input parameter using the data set, storing the generated model in the storage device, performing a search process to search for a value of the input parameter based on the model and a target value of the output parameter, and presenting a second interface that displays a result of the search process , and in the classification process, selecting the reference learning data from among the multiple learning data, selecting target learning data from among the multiple learning data, and if all manufacturing processes of the process in the target learning data are included in the process of the reference learning data and the execution order of all manufacturing processes of the process in the target learning data matches the execution order in the process of the reference learning data, it is determined that the target learning data can be aggregated into the same data set as the reference learning data .

A representative embodiment of the present invention can assist in the construction of a learning dataset. Problems, configurations, and advantages other than those described above will become clear from the explanation of the following examples.

FIG. 1 is a block diagram showing an example of the configuration of a system according to a first embodiment. 1 is a flowchart illustrating an example of a process executed by the device condition search support system according to the first embodiment. 1 is a flowchart illustrating an example of a process executed by the device condition search support system according to the first embodiment. FIG. 2 is a diagram illustrating an example of a screen presented by the apparatus condition search support system according to the first embodiment. 11 is a diagram showing an example of an operation field that the equipment condition search support system of the first embodiment presents when a data input button is operated; FIG. 11 is a diagram showing an example of updating input/output data information by the equipment condition search support system of the first embodiment; FIG. 11 is a diagram showing an example of updating input/output data information by the equipment condition search support system of the first embodiment; FIG. 11 is a diagram showing an example of updating input/output data information by the equipment condition search support system of the first embodiment; FIG. 11 is a diagram showing an example of updating input/output data information by the equipment condition search support system of the first embodiment; FIG. 11 is a diagram showing an example of updating input/output data information by the equipment condition search support system of the first embodiment; FIG. FIG. 11 is a diagram showing an example of a screen presented by the apparatus condition search support system according to the first embodiment when a data set creation button is operated. FIG. 11 is a diagram showing an example of a screen that the equipment condition search support system of the first embodiment presents when a manual creation button is operated. 1 is a flowchart illustrating an example of a process executed by the device condition search support system according to the first embodiment. FIG. 11 is a diagram illustrating an example of a screen presented by the equipment condition search support system according to the first embodiment when a workspace creation button is operated. FIG. 13 is a diagram illustrating an example of a screen presented by the apparatus condition search support system according to the first embodiment when a preprocessing setting button is operated. 13 is a diagram showing an example of a screen presented by the apparatus condition search support system according to the first embodiment when a calculation setting button is operated; FIG. 13 is a diagram showing an example of a screen presented by the apparatus condition search support system according to the first embodiment when a calculation setting button is operated; FIG.

The following describes an embodiment of the present invention with reference to the drawings. However, the present invention should not be interpreted as being limited to the description of the embodiment shown below. It will be easily understood by those skilled in the art that the specific configuration can be changed without departing from the concept or spirit of the present invention.

In the configuration of the invention described below, the same or similar configurations or functions are given the same reference symbols, and duplicate explanations are omitted.

The terms "first," "second," "third," and the like used in this specification are used to identify components and do not necessarily limit the number or order.

The position, size, shape, range, etc. of each component shown in the drawings, etc. may not represent the actual position, size, shape, range, etc., in order to facilitate understanding of the invention. Therefore, the present invention is not limited to the position, size, shape, range, etc. disclosed in the drawings, etc.

Figure 1 is a block diagram showing an example of the system configuration of Example 1.

The system is composed of a device condition search support system 100 and a demonstration experiment system 101. The device condition search support system 100 and the demonstration experiment system 101 are connected to each other directly or via a network.

The user terminal 102 is connected to the display system 111. The user performs various operations via a GUI (Graphical User Interface) that the display system 111 presents to the user terminal 102. Specifically, the user performs input etc. required to execute the functions of the calculation execution unit 142.

The demonstration experiment system 101 performs an experiment using a manufacturing device based on the device conditions input from the device condition search support system 100. The demonstration experiment system 101 is composed of a manufacturing device 150 and a measuring device 151. In the demonstration experiment system 101, a sample 160 to be processed is introduced into the manufacturing device 150, and processing is performed on the sample 160 based on the received device conditions. The manufacturing device 150 outputs the processed sample 160 as a processing result to the measuring device 151. The measuring device 151 measures the processed sample 160 and transmits the measurement result to the device condition search support system 100. The measurement result is registered as output data in the data storage unit 130 by the input/output data input unit 132.

The data acquired as the measurement result is, for example, an image. When feature quantities such as dimensions and size within an image are acquired using measurement software, the results acquired by the measurement software become the measurement result. Note that sensor data acquired by the manufacturing equipment 150 can also be treated as the processing result. In this case, measurement by the measuring device 151 is not necessary, and the sensor data is sent to the equipment condition search support system 100 as the measurement result.

Here, the manufacturing equipment 150 is an equipment that requires optimization of the equipment conditions, such as a semiconductor manufacturing equipment, a film forming equipment, a modeling equipment (such as a 3D printer), a cleaning equipment, a coating equipment, a transporting equipment, a packaging equipment, a machining equipment, and a construction equipment (such as heavy machinery). Specific examples of semiconductor manufacturing equipment include a lithography equipment, a film forming equipment, a pattern processing equipment, an ion implantation equipment, a heating equipment, and a cleaning equipment. The lithography equipment includes an exposure equipment, an electron beam drawing equipment, and an X-ray drawing equipment. The film forming equipment includes, for example, a CVD (Chemical Vapor Deposition), a PVD (Physical Vapor Deposition), an evaporation equipment, a sputtering equipment, and a thermal oxidation equipment. The pattern processing equipment includes, for example, a wet etching equipment, a dry etching equipment, an electron beam processing equipment, and a laser processing equipment. The ion implantation equipment includes a plasma doping equipment and an ion beam doping equipment. Heating devices include, for example, resistance heating devices, lamp heating devices, and laser heating devices. Cleaning devices include liquid cleaning devices and ultrasonic cleaning devices.

The equipment condition search support system 100 searches for equipment conditions to be set for the manufacturing equipment 150. The equipment condition search support system 100 transmits the equipment conditions obtained by the search process to the demonstration experiment system 101. The equipment condition search support system 100 is composed of a calculation system 110 and a display system 111. In the following description, the various processes in the search process are referred to as calculations.

The calculation system 110 and the display system 111 are composed of computers equipped with a processor, memory, a network interface, an input device, and an output device. Note that there is no limit to the number of computers that make up each system.

The calculation system 110 has a calculation unit 120. The display system 111 has a data storage unit 130, an authentication unit 131, an input/output data input unit 132, a learning dataset construction unit 133, a device condition transmission unit 134, and a calculation management unit 135. The calculation management unit 135 has a calculation setting unit 140, a calculation result display unit 141, and a calculation execution unit 142.

When the authentication unit 131 accepts access from a user, it displays a login screen. The login screen is not shown in the figure because it is a general display screen. The authentication unit 131 manages information that associates inputs such as a user name, access password, and group to which the user belongs with accessible areas, and determines whether access is allowed based on that information.

When the login process is complete, the display system 111 displays a screen (see FIG. 5) for inputting data to be used for learning. The input/output data input unit 132 registers and deletes input data and output data, and sets and displays label names and parameter names of data display columns.

Here, the input data is a combination of parameter values that are conditions for controlling the manufacturing equipment 150, and the output data is the processing result of the manufacturing equipment 150. In the following explanation, the pair of input data and output data will be referred to as input/output data. The input/output data is used as data for learning the search model.

When the user has completed inputting input/output data, the display system 111 displays a screen (see Figures 7 and 8) for constructing a learning dataset. The learning dataset construction unit 133 constructs the dataset, manages information associated with the dataset, and displays various information.

When the user has completed construction of a learning dataset, the display system 111 displays a screen for making various settings related to the calculation. The screen includes a screen for creating and managing workspaces (see FIG. 10), a screen for setting preprocessing (see FIG. 11), and a screen for setting a search model (see FIGS. 12A and 12B), as described below. Each functional unit of the calculation management unit 135 reflects settings using the above-mentioned screens, displays the status and results of calculations, issues instructions to execute calculations, and outputs and manages results obtained from the calculation system 110. The calculation management unit 135 may simultaneously execute settings, execute calculations, and display results, or may cause the calculation unit 120 to execute multiple calculations simultaneously.

The data storage unit 130 manages various information handled by the display system 111.

The device condition transmission unit 134 obtains the device conditions from the data storage unit 130 and transmits them to the manufacturing device 150.

The calculation system 110 and the display system 111 may be integrated into a single system. The functional units of the calculation system 110 and the display system 111 may be arranged as desired. For example, the calculation management unit 135 may be included in the calculation system 110. The functional units of the device condition search support system 100 may be multiple functional units combined into one functional unit, or one functional unit may be divided into multiple functional units for each function.

In this embodiment, a system for searching the equipment conditions of the manufacturing equipment 150 is described, but optimization of the equipment conditions to realize the target inspection, observation, or measurement is also important for inspection equipment, observation equipment, or measurement equipment. For example, in observation with an electron microscope, the target image target can be captured with optimal image quality by adjusting equipment conditions such as the focus during imaging, the orientation of the observation surface, the observation position, and the electron beam acceleration voltage. Therefore, the present invention can also be applied to equipment other than manufacturing equipment. Specific examples of equipment to which this system can be applied include electron microscopes, laser microscopes, optical microscopes, film thickness measuring instruments, and three-dimensional dimension measuring instruments.

Figures 2 and 3 are flowcharts illustrating an example of processing executed by the device condition search support system 100 of the first embodiment.

After the user's login process is completed, the device condition search support system 100 starts the process described below.

First, the device condition search support system 100 sets input/output data based on user input via the screen (step S201).

The device condition search support system 100 constructs a learning dataset based on user input via the screen (step S202).

The device condition search support system 100 creates a workspace based on user input via the screen (step S203). At this time, the user selects a dataset to be used for the calculation. The dataset may be selected at the same time as the workspace is created, or may be selected after the workspace is created.

The device condition search support system 100 executes a device condition search process (step S204). In this process, multiple calculations can be set, and each calculation can be executed and the calculation results can be displayed. Details of the device condition search process are explained using FIG. 3.

Datasets, calculation settings, and the execution status of calculations are managed in association with one workspace. This improves the convenience of managing device conditions. Since device conditions are highly dependent on the data set and calculation settings, it is extremely important to simplify the management of these changes and calculation results in order to improve user convenience when searching for device conditions.

The equipment condition search support system 100 transmits an instruction to execute a demonstration experiment, including the equipment conditions, to the demonstration experiment system 101 (step S205). The demonstration experiment system 101 executes a demonstration experiment based on the equipment conditions, and transmits the processing results and the measurement results to the equipment condition search support system 100.

When the device condition search support system 100 receives the processing results and the measurement results, it registers them as output data in the data storage unit 130 (step S206) and ends the process. The output data is registered in association with the transmitted device conditions (input data).

By carrying out the above series of processes, new input/output data is registered in the data storage unit 130. By carrying out the same process again using the new input/output data, new device conditions can be searched for, and a demonstration experiment based on the new device conditions can be carried out. By repeatedly carrying out the process shown in Figure 2, it is possible to significantly improve the accuracy of searching for device conditions that will achieve the user's desired objective.

Before executing the process shown in FIG. 2, the construction of the dataset, the execution of the calculation, the continuation of the demonstration experiment, and the registration of the input/output data can be automated by specifying the method for constructing the dataset and the setting conditions for the calculation in advance. In addition, the device condition search support system 100 can improve user convenience by displaying the progress of various processes to the user.

Next, the device condition search process will be explained using Figure 3.

The device condition search support system 100 performs settings related to preprocessing of the data included in the dataset based on user input via the screen (step S301), and executes preprocessing based on the settings (step S302).

The device condition search support system 100 sets the search model based on the user input via the screen (step S303), executes a learning process based on the settings, and searches for device conditions using the search model (step S304). Here, the search model is, for example, a model that accepts input data as input and predicts output data.

Based on user input via the screen, the device condition search support system 100 sets the post-processing to be performed on the search results (step S305), executes the post-processing based on the settings, and displays the processing results (step S306).

Figure 4 shows an example of a screen presented by the device condition search support system 100 of the first embodiment.

The screen 400 includes an operation button display field 401 and an operation field 402. The operation button display field 401 is a field that displays buttons for selecting operations to be performed using the screen. The operation button display field 401 includes a data input button 410, a dataset creation button 411, and a workspace creation button 412. The operation field 402 is a field for presenting various types of information and setting various types of information. The operation field 402 displays information according to the operation button.

Figure 5 is a diagram showing an example of the operation column 402 that the device condition search support system 100 of the first embodiment presents when the data input button 410 is operated.

When the data input button 410 is pressed, input/output data information 500 and an add button 501 are displayed in the operation column 402. The input/output data information 500 is information in a table format, and stores entries including a condition number 511, a step label 512, a step number 513, an input parameter 514, and an output parameter 515. There is one entry for one condition.

The input and output data for each step included in the processing of the sample 160 are managed in correspondence with the condition number. When learning the relationship between input and output data, it is necessary to understand the corresponding relationship between the input and output of the manufacturing equipment 150, so it is important to manage the data in the format described above.

The fields included in an entry are not limited to those described above. For example, to improve convenience, an entry may include fields that store the name of the registered user, the name of the group to which the user belongs, and the date and time of registration.

Note that the input/output data information 500 is not limited to a table format, but may be in a cell format, sheet format, array format, etc.

Condition number 511 is a field that stores identification information of a condition. Step label 512 is a field that stores a label assigned to a step included in the processing of sample 160. Step number 513 is a field that stores the execution order of steps in the processing of sample 160. Input parameters 514 are a group of fields that store the values of input parameters. Input parameters 514 include one or more fields. Output parameters 515 are a group of fields that store the values of output parameters. Output parameters 515 include one or more fields.

As in the entry with condition number 511 "Xab1", in order to achieve advanced processing, multiple steps are provided in one process, and input parameters are set for each step. Note that output parameters 515 are parameters related to the results of the processing, so one is set for each entry.

If registered input/output data exists, the input/output data information 500 displays an entry corresponding to that input/output data.

By pressing the Add button 501, the user adds an entry to the input/output data information 500, adds a step row to the added entry, and sets values for the step label 512, step number 513, input parameter 514, and output parameter 515 of each row. When the Add button 501 is pressed, the device condition search support system 100 may display a separate input screen and add an entry to the input/output data information 500 based on the input on that screen.

The value set in the input parameter 514 is usually a numerical value, and the manufacturing apparatus 150 has input restrictions such as an upper limit, a lower limit, or an input prohibited range. Therefore, the apparatus condition search support system 100 performs an error judgment on the value entered by the user according to the input restrictions, and if an inappropriate value is entered, it is desirable to highlight the erroneous value or the field into which the value was entered, and also display the reason for the error and notify the user. The value set in the output parameter is also a numerical value, and an upper limit, a lower limit, or an input prohibited range is set according to the sample 160 to be measured. Therefore, the apparatus condition search support system 100 performs an error judgment on the value entered by the user according to the prohibited range, and if an inappropriate value is entered, it is desirable to highlight the erroneous value or the field into which the value was entered, and also display the reason for the error and notify the user.

The device condition search support system 100 may be configured to accept operations to add or delete fields in the input parameters 514 and the output parameters 515, and to change the names of fields. In addition, the number and names of the fields in the input parameters 514 and the output parameters 515 may be set only the first time.

The device condition search support system 100 may also display part of the condition number 511, step label 512, step number 513, input parameters 514, and output parameters 515 on a separate screen.

Note that the condition number 511, step label 512, step number 513, input parameters 514, and output parameters 515 do not have to be managed as a single piece of information.

The device condition search support system 100 classifies input/output data that can be aggregated into the same dataset based on the step labels 512, and can construct a dataset based on the classification results. However, there is input/output data that should be included in the dataset, and input/output data that will make learning difficult if included in the dataset. For example, if the number of input parameters of the dataset increases by including one input/output data in an existing dataset, the dimension of the parameter space formed by the learning data also increases, making it difficult to search for a solution in learning and likely to reduce the accuracy of the search model. For example, there may be a case where the number of input parameters increases to 20 by adding a small amount of data to a dataset with 100 pieces of data and 10 input parameters.

When the dimension of the parameter space is constant, an increase in the number of data is effective in improving the accuracy of the search model. On the other hand, if the dimension of the parameter space increases as the number of data increases, the accuracy of the search model may decrease. Therefore, the increase in the number of parameters due to aggregation into a dataset should be kept as small as possible. It is desirable to keep the increase in the number of parameters at least below the number of data added to the dataset. In addition, an important point in constructing a dataset is to what extent an increase in the number of parameters should be tolerated in relation to the amount of added data.

Aggregating input and output data while considering the trade-off between an increase in the number of data and an increase in the number of parameters requires experience and time, as it is necessary to understand the configuration of steps and parameter types before selecting them.

The device condition search support system 100 of this embodiment visualizes the above-mentioned trade-offs. This allows the user to easily understand the relationship between the above-mentioned trade-offs and to build a high-quality data set in a short period of time.

Here, we will explain how the device condition search support system 100 classifies input/output data. Figures 6A, 6B, 6C, 6D, and 6E are diagrams showing examples of updating the input/output data information 500 by the device condition search support system 100 of Example 1.

The device condition search support system 100 selects input/output data corresponding to two conditions including a plurality of steps, and judges whether the input/output data can be aggregated based on the following first and second criteria. Here, the two selected conditions are the first and second conditions.
(First criterion) All step labels of the first condition are included in the step labels of the second condition.
(Second criterion) The execution order of the step labels in the first condition is the same in the second condition.

When the first and second criteria are met, the device condition search support system 100 determines that the input/output data of the first condition can be aggregated into the input/output data of the second condition. Note that even when the step label included in the first condition and the step label included in the second condition are the same, the first and second criteria are considered to be met.

The device condition search support system 100 automatically classifies input and output data and constructs a dataset using the following procedure.

(Process 1) The device condition search support system 100 selects the condition with the greatest number of steps from the input/output data information 500. The device condition search support system 100 sets the selected condition as the parent condition.

(Process 2) The device condition search support system 100 searches for conditions that satisfy the first and second criteria for the parent conditions, and generates a list of those conditions.

(Process 3) The device condition search support system 100 selects one condition from the list of conditions and sets that condition as the parent condition.

Then, by repeatedly executing (Process 2) and (Process 3), it is possible to classify the input and output data that can be aggregated.

(Process 4) After data classification is complete, the equipment condition search support system 100 builds a dataset by aggregating conditions that are determined to be aggregatable. At this time, the step structure of the conditions included in the dataset is updated based on the parent conditions. Specifically, the equipment condition search support system 100 adds steps included in the parent conditions to the conditions so that the steps have the same structure as the step structure of the parent conditions. Zero is set to the input parameters 514 and output parameters 515 corresponding to the added steps. Even if the conditions are updated as described above, if all the input parameters are zero, the manufacturing equipment 150 usually does not operate, so this can be treated as being synonymous with a state where there are no steps.

Note that parameters that are permitted to be left blank according to the specifications of the manufacturing equipment 150 may be left blank. Note that there are parameters that are difficult to change instantly and take time to reach the specified value, such as the temperature of a liquid medium or heater for regulating the temperature of equipment components. For such parameters, it may be desirable to input the same value as the value of the step prior to the inserted step as the input value for the inserted step, or to input a specific value specified by the user. The user sets the parameter so that the same value as the step prior to the inserted step or a specified value is input.

In the input/output data information 500 shown in FIG. 5, when the condition number "XZ1" is set as the parent condition, the conditions of condition numbers "XZ1", "Xst1", and "Xst2" are aggregated. When the conditions are aggregated to construct a data set, the conditions are updated as shown in FIG. 6A. When the condition number "XY1" is set as the parent condition, the conditions of condition numbers "XY1", "abc1", and "abc2" are aggregated. When the conditions are aggregated to construct a data set, the conditions are updated as shown in FIG. 6B . When the condition number "Xab1" is set as the parent condition, the conditions of condition numbers "Xab1", "Xab2", "Xab3", "Xab4", "abc1", and "abc2" are aggregated. When the conditions are aggregated to construct a data set, the conditions are updated as shown in FIG. 6C. When the condition number "Xab3" is set as the parent condition, the conditions of condition numbers "Xab3", "abc1", and "abc2" are aggregated. When conditions are aggregated to construct a dataset, the conditions are updated as shown in FIG. 6D. When condition number "Xab4" is set as the parent condition, the conditions with condition numbers "Xab4", "abc1", and "abc2" are aggregated. When condition number "abc1" is set as the parent condition, the conditions with condition numbers "abc1" and "abc2" are aggregated. When conditions are aggregated to construct a dataset, the conditions are updated as shown in FIG. 6E.

The conditions for aggregation may consist only of a condition where the input parameter value in the same step has a single non-zero value, and a condition where the parameter for the corresponding step is inserted by aggregation (in other words, the parameter for the corresponding step is not included in the data before aggregation). In this case, if the single non-zero value is X1, it is desirable to set the value of the corresponding parameter to X1 when inserting the step after aggregation. The reason for this is that in a dataset where the parameter value is only X1 and a dataset where the parameter value is both X1 and zero, the data space of the latter dataset is expanded, increasing the time required to search for a solution.

The above-mentioned step insertion is important because a model (e.g., a regression model) that learns the relationship between the input and output of the manufacturing equipment 150 is usually trained using data with the same parameter structure. Also, as mentioned above, manually classifying conditions with different step structures takes time and effort, so incorporating a function that automates or assists in classification and aggregation is important for improving convenience and expanding the number of users.

The above describes the methods for determining which input/output data can be aggregated and the aggregation methods. Below, we will explain examples of input/output data for which aggregation is undesirable. The above aggregation method can also be applied to conditions with different parameters and step structures. For example, if the step labels and step orders of the first and second conditions are different, then in principle they can be aggregated by inserting all of the steps of both into each other. However, if the number of parameters in the first and second conditions is 10 each, the number of parameters after aggregation will increase to a maximum of 20. If the increase in the number of parameters is greater than the increase in the data set due to aggregation, as mentioned above, it becomes difficult to search for a solution. Therefore, aggregation as described above is undesirable.

Fig. 7 is a diagram showing an example of a screen presented by the device condition search support system 100 of the first embodiment when the data set creation button 411 is operated. Fig. 8 is a diagram showing an example of a screen presented by the device condition search support system 100 of the first embodiment when the manual creation button is operated.

When the Create Data Set button 411 is pressed, an Automatic Creation button 420 and a Manual Creation button 421 are newly displayed in the operation button display field 401. Data set information 700 is displayed in the operation field 402.

The dataset information 700 is information in table format, and stores entries including the dataset name 711, parent condition 712, number of conditions 713, number of steps 714, creation date and time 715, and creation method 716. There is one entry for one dataset.

Note that the fields included in an entry are not limited to those described above. For example, an entry may include fields that store the number of parameters for each of the input data and output data.

Parent condition 712 is a field that stores identification information of the parent condition at the time of classification judgment. Number of conditions 713 is a field that stores the number of conditions (input/output data) included in the dataset. Number of steps 714 is a field that stores the number of steps included in the conditions included in the dataset. Creation date and time 715 is a field that stores the date and time when the dataset was created. Creation method 716 is a field that stores the method of creating the dataset. Either "automatic" or "manual" is stored in creation method 716.

The device condition search support system 100 may display a list of data constituting a dataset when a user selects an entry of the dataset information 700. For example, when the entry "XZ 1 -auto-01" is selected, a screen such as that shown in FIG. 6A is displayed.

When the automatic creation button 420 is pressed, the device condition search support system 100 executes the classification process and dataset construction process described above. In FIG. 7, the dataset with the name that includes "auto" is an automatically created dataset.

When the manual creation button 421 is pressed, the device condition search support system 100 presents an operation column 800 shown in FIG. 8 in the operation column 402. The operation column 800 stores entries including parent condition selection 811, data selection 812, condition number 813, step label 814, step number 815, input parameters 816, and output parameters 817. One entry exists for one condition.

Condition number 813, step label 814, step number 815, input parameters 816, and output parameters 817 are the same fields as condition number 511, step label 512, step number 513, input parameters 514, and output parameters 515. Parent condition selection 811 is a field for selecting the condition corresponding to the entry as the parent condition. Data selection 812 is a field for selecting the conditions to be aggregated.

When the parent condition selection 811 of any entry is selected, the device condition search support system 100 executes the classification process described above and highlights the conditions that can be aggregated. As shown in FIG. 8, when the condition "Xab1" is selected as the parent condition, the conditions "abc1", "abc2", "Xab1", "Xab2", "Xab3", and "Xab4" are highlighted. The user refers to the highlighted entries and checks the data selection 812 of the conditions to be aggregated.

As mentioned above, manually classifying conditions takes time and effort, so a function that assists in classification is useful for improving convenience and expanding the number of users.

Note that, although it has been explained that the output parameter value is always set for one condition, there are cases where the output parameter is not measured. Normally, a condition where the output parameter value is not set cannot be used as training data for a model. Therefore, a model is generated by learning the input/output relationship using a condition where all input and output parameter values exist, and the predicted value of a non-existent output parameter is obtained by inputting the input parameter value into the model. The input/output data with the output parameter value complemented can be used as training data. This makes it possible to increase the amount of training data.

However, since the predicted value contains errors resulting from model learning, it is desirable for the user to be able to select whether or not to use the input/output data supplemented by the predicted value. In this case, the device condition search support system 100 makes the predicted value visible by highlighting it, such as by changing the color of the predicted value or the cell in which the predicted value is entered.

Figure 9 is a flowchart illustrating an example of processing executed by the device condition search support system 100 of the first embodiment. Figure 9 illustrates the details of constructing a dataset.

The device condition search support system 100 accepts user input via the screen 400 (step S901).

The device condition search support system 100 creates a dataset based on the user input (step S902).

Specifically, when the automatic creation button 420 is pressed, the device condition search support system 100 automatically classifies the conditions and creates a data set based on the classification results. When the manual creation button 421 is pressed, the device condition search support system 100 classifies the conditions based on the parent conditions specified by the user and creates a data set based on the conditions selected by the user.

The device condition search support system 100 performs prediction of the output parameter value for the condition for which the output parameter value has not been set (step S903). The device condition search support system 100 sets the predicted value to the output parameter.

To realize the processing of step S903, an operation button for executing prediction of the value of the output parameter is provided in the operation button display column 401 or the operation column 402. The user selects a dataset and presses the operation button to execute the processing of step S903. Note that a dataset does not have to be specified. In this case, the processing is executed for all datasets.

Note that if there are no conditions for which the value of the output parameter is not set, the processing of step S903 can be omitted.

The user may be allowed to choose whether or not to include input and output data, including predicted values, in the dataset.

Figure 10 is a diagram showing an example of a screen that the device condition search support system 100 of the first embodiment presents when the workspace creation button 412 is operated.

When the workspace creation button 412 is pressed, a new creation button 430, a copy button 431, a pre-processing setting button 432, a calculation setting button 433, and a result display button 434 are displayed in the operation button display field 401. The workspace information 1000 is displayed in the operation field 402.

The workspace information 1000 is information in a table format, and stores entries including a workspace name 1011, a dataset name 1012, a parent condition 1013, the number of conditions 1014, the number of steps 1015, creation date and time 1016, calculation start date and time 1017, and calculation status 1018. One entry exists for one workspace.

Note that the fields included in an entry are not limited to those described above. For example, an entry may include fields that store the number of parameters for each of the input data and output data.

Dataset name 1012, parent condition 1013, number of conditions 1014, number of steps 1015, and creation date and time 1016 are the same fields as dataset name 711, parent condition 712, number of conditions 713, number of steps 714, and creation date and time 715. Workspace name 1011 is a field that stores the name of the workspace. The name of the workspace may be set automatically, or may be set manually in the field by the user. Calculation start date and time 1017 is a field that stores the start date and time of the search process. Calculation status 1018 is a field that stores the execution status of the calculation in the search process.

When the Create New button 430 is pressed, the device condition search support system 100 presents a dataset selection screen (not shown), and when it receives the dataset selection result from the user, it registers an entry in the workspace information 1000.

When the user selects any entry in the workspace information 1000 and presses the copy button 431, the device condition search support system 100 adds an entry to the workspace information 1000 and copies the contents of the selected entry to that entry. Note that the workspace name 1011 is set to a value different from that of the selected entry.

Because the search for device conditions is highly dependent on the data set and calculation settings, in order to improve user convenience when searching for device conditions, it is important to simplify the changes to the data set and calculation settings, as well as the management of the results of the changes. Therefore, as shown in Figure 10, user convenience can be improved by managing the data set, calculation settings, and the execution status of the calculation for each workspace.

When the pre-processing setting button 432, the calculation setting button 433, and the result display button 434 are pressed, a screen corresponding to each button is displayed. Details of the screens will be described later. Note that if the search process is not complete, the device condition search support system 100 does not accept operation of the result display button 434, or performs control so that the screen is not displayed even if the result display button 434 is operated.

Figure 11 is a diagram showing an example of a screen that the device condition search support system 100 of Example 1 presents when the pre-processing setting button 432 is operated.

Screen 1100 includes a display section 1101, setting sections 1102, 1103, 1104, and an operation section 1105.

The display field 1101 is a field that displays information about the workspace being used. The setting field 1102 is a field for setting target values for output parameters. The user sets the target values for the output parameters in the setting table 1120.

The setting field 1103 is a field for setting the search range of the input parameters. In the setting table 1130, the user sets the maximum and minimum values of the input parameters, as well as the number of values to be added during the search that are not between the maximum and minimum values. The reason for adding values to the search range is that some input parameters, such as the gas inflow rate system, have a control range in which the inflow is zero when completely closed, and non-zero lower and upper limits for inflow control.

The setting field 1104 is a field for setting input parameters whose values change discretely during a search. The user sets a discrete change pattern for any input parameter in the setting table 1140. This makes it possible to search for values of input parameters that can only take discrete values or that are set only by discrete values.

Settings fields 1102, 1103, and 1104 include a save button for saving the settings and a load button for loading the saved settings.

The operation column 1105 is a column for controlling input to the setting columns 1102, 1103, and 1104, and the execution of preprocessing.

Edit button 1150 is a button for enabling input to setting fields 1102, 1103, and 1104. When edit button 1150 is pressed, the execution state of the calculation is initialized.

The execute button 1151 is a button for transitioning to the execution phase of the calculation related to pre-processing. When the execute button 1151 is pressed, control is performed so that input to the setting fields 1102, 1103, and 1104 is not accepted. The results of the calculation are managed in correspondence with various settings, and this is to avoid confusion in the correspondence with the results of the calculation when various settings are changed after the calculation is executed. In particular, when executing multiple calculations, if the settings are changed in the middle of the calculation, it may cause an error in the calculation, such as stopping the calculation, so it is necessary to switch between editing the settings and executing the calculation.

When the execute button 1151 is pressed, the device condition search support system 100 may check the setting contents of the setting field 1102 and determine whether or not the correct value has been set. This can prevent errors from occurring.

The operation table 1155 displays the name of the preprocessing operation, an execution button, and the execution status of the operation. In this embodiment, each operation is executed in order from the top. The execution button for each operation is not enabled until the previous operation is completed. For example, the execution button for "Preprocess2" is enabled after "Preprocess1" is completed. This allows the user to visually grasp the execution status of the operations, improving convenience.

The log column 1156 displays a log showing the execution status or results of the calculation.

Result folder button 1152 is a button for referencing the folder in which output data etc. generated by executing a calculation is saved. Transition button 1153 is a button for transitioning to the phase before pre-processing setting. For example, when transition button 1153 is pressed, transition to the state in FIG. 10 occurs. Transition button 1154 is a button for transitioning to the calculation setting phase. For example, when transition button 1154 is pressed, transition to the state in FIG. 12 occurs. However, transition button 1154 is not enabled until all calculations are completed.

Figures 12A and 12B are diagrams showing an example of a screen that the device condition search support system 100 of Example 1 presents when the calculation setting button 433 is operated.

Screen 1200 includes setting columns 1201, 1202, an operation column 1203, and a display column 1204.

The setting field 1201 is a field for configuring settings related to the learning of the search model and the calculations related to the search of device conditions.

The user selects the model type to be used and the search method for device conditions using the model in the model selection table 1210. Note that the model selection table 1210 displays the available search methods according to the selected model type in a selectable format.

The model format can be selected from simple regression, support vector regression, regression tree, regression using neural network, etc. The search method can be selected from random search, grid search, gradient method, etc.

The user sets the values of parameters for controlling the search for device conditions in parameter setting tables 1211 and 1212. Parameter setting table 1211 is a table for setting parameters that depend on the model format and search method, and parameter setting table 1212 is a table for setting parameters that do not depend on the model format and search method.

When "Kernel Ridge Regression" is selected as the model format and "Random" is selected as the search method, parameter setting table 1211 and parameter setting table 1212 display parameters as shown in FIG. 12A.

Cnum is a parameter that adjusts the number of input parameters that are changed at one time during a search. searchN is a parameter that adjusts the number of conditions to be searched. Nprint is a parameter that adjusts the display interval of intermediate results. When the same number of search trials as the value set in Nprint have been performed, the device condition search support system 100 displays the processing conditions obtained up to that point. Nkf is a parameter that adjusts the number of data divisions in cross-validation.

The search cost table 1213 displays the recommended number of searches and an estimate of the calculation time. The recommended calculation time may also be displayed. The number of searches and the calculation time are calculated based on the settings in the model selection table 1210 and the parameter setting tables 1211 and 1212.

The setting field 1202 is a field for configuring hyperparameters.

When the user wants to perform a search for hyperparameters, the user presses the search button 1220. The device condition search support system 100 searches for hyperparameters and displays the search results in the hyperparameter table 1221. When the user wants to manually set the hyperparameters, the user presses the setting button 1222 and sets the values in the setting table 1223.

An operation column 1203 is a column for controlling input to the setting columns 1201 and 1202 , as well as the learning of the search model and the search for device conditions.

Edit button 1230 is a button for enabling input to setting fields 1201 and 1202. In addition, when edit button 1230 is pressed, the execution state of the calculation is initialized.

The execute button 1231 is a button for transitioning to the execution phase of the calculation. When the execute button 1231 is pressed, control is performed so that input to the setting fields 1201 and 1202 is not accepted.

Calculation table 1235 and log column 1236 are similar columns to calculation table 1155 and log column 1156.

Result folder button 1232 is a button for referencing a folder in which output data generated by executing a calculation, etc., is saved. Transition button 1233 is a button for transitioning to a phase before the calculation was set. For example, when transition button 1233 is pressed, a transition to the state shown in FIG. 11 occurs. Transition button 1234 is a button for transitioning to a result reference phase. For example, when transition button 1234 is pressed, the results are displayed in display field 1204. However, transition button 1234 is not enabled until all calculations are completed.

Display field 1204 is a field for displaying the device conditions obtained by the calculation. Multiple device conditions are displayed in display field 1204.

As described above, the equipment condition search support system 100 of Example 1 can support various inputs for searching equipment conditions for manufacturing equipment. In particular, it is characterized by supporting the construction of a learning dataset, which can reduce the time and effort required to optimize equipment conditions and improve efficiency. In addition, since the equipment condition search support system 100 supports various inputs, it can also be used by engineers who do not have specialized knowledge.

The second embodiment differs from the first embodiment in that the device condition search support system 100 is used by multiple users.

The multiple user terminals 102 may each be connected to the device condition search support system 100 via a different network. For example, they may be connected via the Internet or via an intranet. A firewall and an authentication server are installed between the user terminals 102 and the device condition search support system 100, and user authentication and access control are performed.

A user can operate the user terminal 102 to access the device condition search support system 100 and input input data and output data, create a dataset, create a workspace, set calculations, and obtain device conditions. The user can also send device conditions to the demonstration experiment system 101 and obtain the results of the demonstration experiment. The user can also register the demonstration experiment results together with the device conditions in the device condition search support system 100. By registering the demonstration experiment results as new learning data, the search accuracy for device conditions can be improved.

By registering training data, multiple users can share equipment conditions, data sets, and demonstration experiment results, allowing multiple users to collaborate in exploring equipment conditions.

The present invention is not limited to the above-described embodiments, but includes various modified examples. For example, the above-described embodiments are provided to explain the present invention in detail, and are not necessarily limited to those including all of the described configurations. In addition, it is possible to add, delete, or replace part of the configuration of each embodiment with another configuration.

The above-mentioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole by hardware, for example by designing them as integrated circuits. The present invention can also be realized by software program code that realizes the functions of the embodiments. In this case, a storage medium on which the program code is recorded is provided to a computer, and a processor included in the computer reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium realizes the functions of the above-mentioned embodiments, and the program code itself and the storage medium on which it is stored constitute the present invention. Examples of storage media for supplying such program code include flexible disks, CD-ROMs, DVD-ROMs, hard disks, SSDs (Solid State Drives), optical disks, magneto-optical disks, CD-Rs, magnetic tapes, non-volatile memory cards, and ROMs.

In addition, the program code that realizes the functions described in this embodiment can be implemented in a wide range of program or script languages, such as assembler, C/C++, perl, Shell, PHP, Python, Java (registered trademark), etc.

Furthermore, the program code of the software that realizes the functions of the embodiment may be distributed over a network and stored in a storage means such as a computer's hard disk or memory, or in a storage medium such as a CD-RW or CD-R, and the processor of the computer may read and execute the program code stored in the storage means or storage medium.

In the above examples, the control lines and information lines are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. All components may be interconnected.

100 Equipment condition search support system 101 Demonstration experiment system 102 User terminal 110 Calculation system 111 Display system 120 Calculation unit 130 Data storage unit 131 Authentication unit 132 Input/output data input unit 133 Learning data set construction unit 134 Equipment condition transmission unit 135 Calculation management unit 140 Calculation setting unit 141 Calculation result display unit 142 Calculation execution unit 150 Manufacturing equipment 151 Measuring instrument 160 Sample

Claims (8)

A computer system for supporting a search for equipment conditions for controlling a manufacturing equipment that performs a process, comprising:
At least one computer having a computing device, a storage device connected to the computing device, and a connection interface connected to the computing device;
The process includes at least one manufacturing step,
The computer system comprises:
presenting a first interface for receiving input of learning data configured of a combination of values of input parameters, which are the device conditions, and values of output parameters, which are results of the processing;
storing the learning data input via the first interface in the storage device;
Analyzing a structure of a manufacturing process of the process in the reference training data and a structure of a manufacturing process of the process in other training data, and performing a classification process to identify the training data that can be aggregated into the same data set as the reference training data based on a result of the analysis;
generating a dataset by aggregating the training data based on the results of the classification process;
Using the data set, a learning process is performed for a model that predicts the value of the output parameter from the value of the input parameter, and the generated model is stored in the storage device;
performing a search process to search for values of the input parameters based on the model and the target values of the output parameters;
presenting a second interface displaying results of the search process ;
In the classification process,
selecting the reference learning data from among the plurality of learning data;
Selecting target training data from among the plurality of training data;
A computer system characterized by determining that the target learning data can be aggregated into the same data set as the reference learning data when all manufacturing processes of the process in the target learning data are included in the process of the reference learning data and the execution order of all manufacturing processes of the process in the target learning data matches the execution order of the process in the reference learning data. 2. The computer system of claim 1,
A computer system characterized in that, when generating a dataset using the learning data that can be aggregated into the reference learning data, the aggregatable learning data is shaped so that a structure of a manufacturing process of the processing in the aggregatable learning data is identical to a structure of the manufacturing process of the processing in the reference learning data. 2. The computer system of claim 1,
and a computer system presenting a third interface for specifying the reference training data. 2. The computer system of claim 1,
A computer system comprising: a fourth interface for selecting the training data to be aggregated; A method for supporting a search for equipment conditions for controlling a manufacturing equipment performing a process, the method being executed by a computer system, comprising the steps of:
The computer system includes at least one computer having a computing device, a storage device connected to the computing device, and a connection interface connected to the computing device;
The process includes at least one manufacturing step,
The method for supporting search for device conditions comprises:
a first step in which the at least one computer presents a first interface for receiving input of learning data configured of a combination of values of input parameters, which are the device conditions, and values of output parameters, which are results of the processing;
a second step in which the at least one computer stores the learning data input via the first interface in the storage device;
a third step in which the at least one computer analyzes a structure of a manufacturing process of the process in reference training data and a structure of a manufacturing process of the process in other training data, and executes a classification process to identify the training data that can be aggregated into the same data set as the reference training data based on a result of the analysis;
a fourth step of generating a dataset by aggregating the training data based on a result of the classification process by the at least one computer;
a fifth step in which the at least one computer executes a learning process of a model that predicts a value of the output parameter from a value of the input parameter using the data set, and stores the generated model in the storage device;
a sixth step of performing, by the at least one computer, a search process for searching values of the input parameters based on the model and target values of the output parameters;
and a seventh step of the at least one computing device presenting a second interface displaying results of the search process;
The third step includes:
The at least one computer selects the reference training data from among a plurality of the training data;
selecting target training data from among the plurality of training data by the at least one computer;
a step of determining, by the at least one computer, that the target learning data can be aggregated into the same data set as the reference learning data when all manufacturing processes of the process in the target learning data are included in the process of the reference learning data and the execution order of all manufacturing processes of the process in the target learning data matches the execution order of the process in the reference learning data. 6. The method for supporting a search for device conditions according to claim 5,
The fourth step is a method for supporting search for equipment conditions, characterized in that it includes a step of at least one computer shaping the aggregatable learning data so that a structure of a manufacturing process of the processing in the aggregatable learning data is identical to a structure of the manufacturing process of the processing in the reference learning data. 6. The method for supporting a search for device conditions according to claim 5,
The method for supporting a search for device conditions, wherein the third step includes a step of the at least one computer presenting a third interface for designating the reference learning data. 6. The method for supporting a search for device conditions according to claim 5,
A search support method for device conditions, wherein the fourth step includes a step of the at least one computer presenting a fourth interface for selecting the learning data to be aggregated.

Images